如何用MongoDB聚合构建依赖关系的嵌套层级结构？

需要在MongoDB的relations集合中，针对指定文档生成完整的自上而下依赖层级。集合中文档的dependentOn字段为Object类型，内部按分类（如metrics、measures）存储依赖的_id数组；终止节点无dependentOn字段，仅包含反向关联的usedBy字段。

示例文档 #

// Doc 1:
{
  "_id": "p_commerce_metrics_dtst_v1",
  "category": "dataset",
  "dependentOn": {
    "metrics": [
      "net_booking_count_v1"
    ]
  }
}
// Doc 2:
{
  "_id": "net_booking_count_v1",
  "category": "metric",
  "dependentOn": {
    "metrics": [
      "cancelled_booking_count_v1",
      "gross_booking_count_v1"
    ]
  }
}
// Doc 3:
{
  "_id": "cancelled_booking_count_v1",
  "category": "metric",
  "dependentOn": {
    "measures": [
      "hb_cancel_measure_v1"
    ]
  }
}
// Doc 4:
{
  "_id": "gross_booking_count_v1",
  "category": "metric",
  "dependentOn": {
    "measures": [
      "hb_booking_measure_v1"
    ]
  }
}
// Doc 5（无依赖，终止节点）:
{
  "_id": "hb_cancel_measure_v1",
  "category": "measure",
  "usedBy": {
    "metrics": [
      "cancelled_booking_count_v1",
      "more_metrics"
    ]
  }
}
// Doc 6（无依赖，终止节点）:
{
  "_id": "hb_booking_measure_v1",
  "category": "measure",
  "usedBy": {
    "metrics": [
      "gross_booking_count_v1",
      "more_metrics"
    ]
  }
}

期望输出 #

获取p_commerce_metrics_dtst_v1的完整依赖层级：

[
  {
    "name_version": "p_commerce_metrics_dtst_v1",
    "category": "dataset",
    "dependent_on": {
      "metrics": [
        "net_booking_count_v1",
        "cancelled_booking_count_v1",
        "gross_booking_count_v1"
      ],
      "measures": [
        "hb_cancel_measure_v1",
        "hb_booking_measure_v1"
      ]
    }
  }
]

尝试的无效查询 #

[
    { $match: {
        _id: "p_commerce_metrics_dtst_v1"
    }},
    { $graphLookup: {
        from: "relations",
        startWith: "$dependentOn",
        connectFromField: "dependentOn",
        connectToField: "_id",
        depthField: "depth",
        as: "dependentOn"
    }}
]

正确的聚合实现方案 #

原查询的问题在于：$graphLookup的startWith和connectFromField需要传入可匹配_id的单一值或数组，而非Object类型的dependentOn。以下是修正后的聚合流程：

[
  // 匹配目标文档
  { $match: { _id: "p_commerce_metrics_dtst_v1" } },
  // 提取初始依赖的所有_id数组，同时保留原文档字段
  {
    $project: {
      category: 1,
      initialDependencies: {
        $reduce: {
          input: { $objectToArray: "$dependentOn" },
          initialValue: [],
          in: { $concatArrays: ["$$value", "$$this.v"] }
        }
      },
      initialDepsByType: "$dependentOn"
    }
  },
  // 递归查询所有依赖节点
  {
    $graphLookup: {
      from: "relations",
      startWith: "$initialDependencies",
      connectFromField: {
        $reduce: {
          input: { $ifNull: [{ $objectToArray: "$dependentOn" }, []] },
          initialValue: [],
          in: { $concatArrays: ["$$value", "$$this.v"] }
        }
      },
      connectToField: "_id",
      as: "allDependencies"
    }
  },
  // 合并所有依赖的分类信息，并整理结构
  {
    $project: {
      name_version: "$_id",
      category: 1,
      dependent_on: {
        $mergeObjects: [
          "$initialDepsByType",
          {
            $arrayToObject: {
              $map: {
                input: "$allDependencies",
                as: "dep",
                in: {
                  $arrayToObject: {
                    $map: {
                      input: { $ifNull: [{ $objectToArray: "$$dep.dependentOn" }, []] },
                      as: "type",
                      in: {
                        k: "$$type.k",
                        v: "$$type.v"
                      }
                    }
                  }
                }
              }
            }
          }
        ]
      }
    }
  },
  // 对每个分类的数组去重并合并
  {
    $set: {
      "dependent_on": {
        $arrayToObject: {
          $map: {
            input: { $objectToArray: "$dependent_on" },
            as: "entry",
            in: {
              k: "$$entry.k",
              v: { $reduce: {
                  input: "$$entry.v",
                  initialValue: [],
                  in: { $setUnion: ["$$value", "$$this"] }
                }
              }
            }
          }
        }
      }
    }
  },
  // 移除多余字段
  { $unset: ["_id"] }
]

关键步骤说明 #

1. 提取初始依赖：用$objectToArray将dependentOn转为键值对数组，再通过$reduce合并所有分类的依赖ID，作为$graphLookup的起始查询值。
2. 递归查询依赖：connectFromField动态提取每个节点的dependentOn中的所有ID，实现层级递归查询。
3. 合并分类结构：用$mergeObjects合并初始依赖和递归查询到的依赖，再通过$arrayToObject和$setUnion整理去重，生成按分类分组的依赖列表。
4. 字段重命名与清理：将_id改为name_version，移除多余字段，匹配期望输出格式。

内容的提问来源于stack exchange，提问作者Prabhjot